> For the complete documentation index, see [llms.txt](https://docs.humblytics.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.humblytics.com/split-testing-overview/deciding-how-long-to-run-an-a-b-test.md).

# Deciding How Long to Run an A/B Test

When to stop a test, and why stopping too early gives you wrong answers.

A test is done when it has **enough visitors** and has **run long enough**, not when the numbers first look good. Use the [A/B Test Planner](/split-testing-overview/using-the-humblytics-a-b-sample-size-calculator.md) to get your visitor target and estimated duration before you launch.

## Recommended run length by goal

Humblytics recommends a minimum and maximum run time for each goal:

| Goal                                   | Recommended run | Suggested significance |
| -------------------------------------- | --------------- | ---------------------- |
| Form Submissions, Click-through Rate   | 7–30 days       | 95%                    |
| Destination Page, External Destination | 14–30 days      | 95%                    |
| Revenue                                | 14–30 days      | 97%                    |
| Visit Time, Bounce Rate                | 3–30 days       | 90%                    |

Revenue tests run longer and aim higher because purchases are rarer and more costly to get wrong.

## When to stop

Stop and pick a winner when the test is **Ready to Call**:

1. Confidence has reached your target.
2. At least 80% of the needed sample is in.
3. It has run for the minimum days for its goal.

Stop without a winner when you've passed the recommended maximum and the verdict is **No significant difference detected** or still **Inconclusive**. The change probably doesn't matter much. Test something bolder.

## Rules of thumb

* **Run whole weeks.** People behave differently on weekends, so a test that runs Monday to Friday can mislead you.
* **Don't stop at the first good number.** Results swing a lot early on. The more often you check and stop, the more likely you pick a false winner.
* **Don't change the pages mid-test.** Editing a variant, or running a big campaign to one page only, mixes old and new results.
* **Low traffic? Test bigger changes.** A new headline and offer is easier to detect than a new button color.

Keep an eye on the **Test Health** panel on the results page. It shows sample progress and days running against these targets. [How to read results](/split-testing-overview/how-to-analyze-split-test-data.md).